OCaml 5.0.0 is out!

We have the pleasure of celebrating the birthdays of Jane Austen and Arthur C. Clarke by announcing the release of OCaml version 5.0.0.

The highlight of this new major version of OCaml is the long-awaited runtime support for shared memory parallelism and effect handlers.

This multicore support is the culmination of more than 8 years of effort, and required a full rewrite of the OCaml runtime environment. Consequently, OCaml 5.0.0 is expected to be a more experimental version of OCaml than the usual OCaml releases.

In this release, the native compiler only supports the x86-64 and arm64 architectures. In terms of operating systems, Linux, the BSDs, macOS and mingw64 on Windows are supported.

We are planning to restore the support for most previously supported architectures and operating systems, and address remaining known issues over the course of the next year.

Don’t let those words of caution hinder your experimentation with OCaml 5:
OCaml 5 as a language is fully compatible with OCaml 4 down to the performance characteristics of your programs. In other words, any code that works with OCaml 4 should work the same with OCaml 5.

The currently known exceptions to this rule are:

  • the removal of many long-deprecated functions and modules
  • changes to the internal runtime API
  • the performance of ephemerons is currently (and temporarily) strongly degraded.

Please report any unexpected behaviours on the OCaml issue tracker.

During the maturing period for OCaml 5, the OCaml 4.14 branch will be maintained by backporting bug fixes from the OCaml 5 branch at regular intervals. The next release on the OCaml 4.14 branch, OCaml 4.14.1, should soon follow this release.

The full list of changes can be found in the changelog below.

Installation Instructions

The base compiler can be installed as an opam switch with the following commands:

opam update
opam switch create 5.0.0

It might be also interesting to check the new support for parallelism by installing the domainslib library with

opam install domainslib

The source code for the release candidate is also directly available on:

Fine-Tuned Compiler Configuration

If you want to tweak the configuration of the compiler, you can switch to the option variant with:

opam update
opam switch create <switch_name> ocaml-variants.5.0.0+options <option_list>

where <option_list> is a comma-separated list of ocaml-option-* packages. For instance, for a flambda and no-flat-float-array switch:

opam switch create 5.0.0+flambda+nffa ocaml-variants.5.0.0+options ocaml-option-flambda ocaml-option-no-flat-float-array

The command line above is slightly more complicated for opam versions before 2.1:

opam update
opam switch create <switch_name> --packages=ocaml-variants.5.0.0+options,<option_list>

In both cases, all available options can be listed with opam search ocaml-option.

Changelog for OCaml 5.0.0 (16 December 2022)

  • #10831: Multicore OCaml
    (Enguerrand Decorne, Stephen Dolan, Tom Kelly, Sadiq Jaffer,
    Anil Madhavapeddy, Sudha Parimala, KC Sivaramakrishnan,
    Leo White, the Tarides multicore team,
    review by Florian Angeletti, Damien Doligez, Xavier Leroy,
    Guillaume Munch-Maccagnoni, Olivier Nicole, NicolĂĄs Ojeda BĂ€r,
    Gabriel Scherer, the OCaml core development team, and many
    other valued reviewers.)

Runtime system:

  • #11400: Runtime events counters fixes
    Fixes mismatch between OCaml and C APIs, removes events from 4.x that
    are not present in the 5.0 GC and adds some missing probes.
    (Sadiq Jaffer, review by Gabriel Scherer, Florian Angeletti)

  • #11368: Runtime events buffer size OCAMLRUNPARAMS fix
    The runtime events buffer size can now be set via the ‘e’ OCAMLRUNPARAM.
    This is previously mistakenly enabled/disabled tracing instead.
    (Sadiq Jaffer, review by KC Sivaramakrishnan, David Allsopp, Damien Doligez)

  • #11308: Add environment variable to preserve runtime_events after exit
    If the environment variable OCAML_RUNTIME_EVENTS_PRESERVE exists then the
    runtime will not remove the runtime events ring buffers at exit. This
    makes tracing very short running programs more reliable.
    (Sadiq Jaffer, review by KC Sivaramakrishnan)

  • #10964: Ring-buffer based runtime tracing (runtime_events)
    Runtime_events is a very low overhead runtime tracing system designed for
    continuous monitoring of OCaml applications.
    (Sadiq Jaffer, review by Anil Madhavapeddy, Enguerrand Decorne,
    Richard Warburton, Gabriel Scherer, Sabine Schmaltz, Florian Angeletti,
    Patrick Ferris, Tom Kelly)

  • (breaking change) #10723: do not use -flat-namespace linking for macOS.
    (Carlo Cabrera, review by Damien Doligez)

  • (breaking change) #10863, #10933: Remove support for old, unprefixed C runtime function names
    such as alloc. The new names prefixed with caml_ must be used instead,
    such as caml_alloc. Consequently, it is no longer needed to define
    CAML_NAME_SPACE to avoid bringing unprefixed names into scope: this is now
    the default behavior.
    (NicolĂĄs Ojeda BĂ€r, review by Xavier Leroy)

  • #10902: Do not register empty code fragments in natdynlink.
    (David Allsopp, review by Xavier Leroy and Damien Doligez)

  • #10965: caml_fatal_error_hook, GC timing hooks, and
    caml_scan_roots_hook are now atomic variables. Restore GC timing
    hooks in multicore.
    (Guillaume Munch-Maccagnoni, review by Enguerrand Decorne, Xavier
    Leroy, Gabriel Scherer, and KC Sivaramakrishnan)

  • #11209: Add a public and thread-safe timing hook running at domain
    termination, after this domain has stopped running any OCaml code:
    caml_domain_terminated_hook. This can be useful for implementing
    domain-local state in C.
    (Guillaume Munch-Maccagnoni, review by Xavier Leroy and Gabriel

  • #10875, #11731: Add option to allocate fiber stacks and sigaltstacks with
    mmap(MAP_STACK) instead of malloc. This is exposed via a configure
    –enable-mmap-map-stack option, and is enabled by default on OpenBSD where
    it is mandatory.
    (Anil Madhavapeddy, review by Gabriel Scherer, Tom Kelly,
    Michael Hendricks and KC Sivaramakrishnan).

  • #10950: Do not use mmap to allocate Caml_state.
    In order to reduce virtual memory usage, we dynamically allocate
    the domain_state structure.
    (Enguerrand Decorne, KC Sivaramakrishnan and Tom Kelly,
    review by Anil Madhavapeddy and Gabriel Scherer)

  • #11010: Use strerror_r for reentrant error string conversion.
    (Anil Madhavapeddy and Xavier Leroy, review by same and Edwin Török)

  • #11002, #11066, #11086: Do not use Begin_roots/End_roots macros in
    the runtime system. Also fix a missing root registration in the
    implementation of Unix.write on Windows.
    (NicolĂĄs Ojeda BĂ€r, Daniel BĂŒnzli and Antonin DĂ©cimo,
    review by Xavier Leroy and David Allsopp)

  • #11022: Track GC work for all managed bigarray allocations
    (Stephen Dolan, report by Andrew Hunter, review by Damien Doligez)

  • #10802: Use 4.12 value macros and helpers in C code
    (Antonin DĂ©cimo, review by Gabriel Scherer)

  • #11105: Fix handling of fiber stack cache with multiple domains
    (Jon Ludlam, KC Sivaramakrishnan and Tom Kelly)

  • #11054: Respect user provided maximum stack space
    Make sure the stack we initially request is sized accordingly to
    the user provided settings. tmc/stack_space is also updated by
    this PR in order to account for this change.
    (Enguerrand Decorne, report by Jon Ludlam,
    review by Tom Kelly, KC Sivaramakrishnan and Gabriel Scherer)

  • #11238: Increase the default limit for the stack size to 128 Mi words,
    i.e. 1 Gib for 64-bit platforms and 512 Mib for 32-bit platforms.
    (Xavier Leroy, review by SĂ©bastien Hinderer)

  • (breaking change) #11295: An ISO C 2011 compliant compiler, including full support for atomic
    types, is now required to build the OCaml runtime system.
    (Xavier Leroy, review by David Allsopp and SĂ©bastien Hinderer)
  • #10915, #11039, #11057, #11095, #11190: Implement quality treatment for
    asynchronous actions in multicore. Reimplement the old behaviour of
    caml_process_pending* for multicore.
    (Guillaume Munch-Maccagnoni, review by Sadiq Jaffer and Gabriel Scherer)

  • #11303: Ensure that GC is not invoked from bounds check failures
    (Stephen Dolan, review by Sadiq Jaffer and Xavier Leroy)

  • #11304: Fix data race on Windows file descriptors
    (Olivier Nicole and Xavier Leroy, review by Xavier Leroy, David Allsopp,
    and Sadiq Jaffer)

  • (breaking change) #11337: pass ‘flags’ metadata to root scanners, to optimize stack
    scanning in the bytecode interpreter.
    Changes the interface of user-provided root-scanning hooks.
    (Gabriel Scherer, review by Xavier Leroy,
    Guillaume Munch-Maccagnoni, Sadiq Jaffer and Tom Kelly)
  • #11144: Restore frame-pointers support for amd64
    (Fabrice Buoro, review by Frederic Bour and KC Sivaramakrishnan)
  • (breaking change) #11255: in the C interface, &Field(v, i) now has type volatile value *
    instead of value * in OCaml 4. This makes the memory model
    for mixed OCaml/C code better defined, but can cause warnings or type
    errors in user C code.
    (KC Sivaramakrishnan, review by Xavier Leroy, Gabriel Scherer and
    Guillaume Munch-Maccagnoni, additional discussions with Stephen
    Dolan and Luc Maranget)
  • #5299, #4787, #11138, #11272, #11506: To help debugging, Caml_state
    now dynamically checks that the domain lock is held, and fails
    otherwise (with a fatal error at most entry points of the C API, or
    systematically in debug mode). A new variable Caml_state_opt is
    introduced, and is NULL when the domain lock is not held. This
    allows to test from C code if the current thread holds the lock of
    its domain.
    (Guillaume Munch-Maccagnoni, review by Florian Angeletti, Damien
    Doligez, Sadiq Jaffer, Xavier Leroy, and Gabriel Scherer)

  • #11223: The serialization format of custom blocks changed in 4.08,
    but the deserializer would still support the pre-4.08 format. OCaml
    5.x removed support for this old format; provide a clear error message
    in this case.
    (Hugo Heuzard, review by Gabriel Scherer)

  • #11504, #11522: Use static allocation in caml_make_float_vect in
    no-flat-float-array mode, it’s more efficient and avoids a a race condition
    (Xavier Leroy, report by Guillaume Munch-Maccagnoni, review by David Allsopp)

  • #11652: Fix benign off-by-one error in Windows implementation of caml_mem_map.
    (David Allsopp, review by Gabriel Scherer)

  • #11669, #11704: Fix construction of Effect.Unhandled exceptions in the
    bytecode interpreter.
    (David Allsopp and Xavier Leroy, report by Samuel Hym, review by Xavier Leroy
    and Gabriel Scherer)

Code generation and optimizations:

  • #10972: ARM64 multicore support: OCaml & C stack separation;
    dynamic stack size checks; fiber and effects support.
    (Tom Kelly and Xavier Leroy, review by KC Sivaramakrishnan, Xavier Leroy
    Guillaume Munch-Maccagnoni, Eduardo Rafael, Stephen Dolan and
    Gabriel Scherer)
  • (breaking change) #10845 Emit frametable size on amd64 BSD (OpenBSD, FreeBSD, NetBSD) systems
    (emitted for Linux in #8805)
    (Hannes Mehnert, review by NicolĂĄs Ojeda BĂ€r)

Standard library:

  • #10742: Use LXM as the pseudo-random number generator for module Random.
    Add Random.State.split and Random.split to “split” a PRNG off
    another PRNG.
    (Xavier Leroy, review by Gabriel Scherer and Hugo Heuzard)
  • (breaking change) #10867, #11345: Remove deprecated values: Array.create, Array.make_float,
    Array.create_matrix, Bytes.uppercase, Bytes.lowercase, Bytes.capitalize,
    Bytes.uncapitalize, Char.lowercase, Char.uppercase, Filename.temp_dir_name,
    Int32.format, Int64.format, Nativeint.format, Format.bprintf, Format.kprintf,
    Format.pp_get_all_formatter_output_functions, Format.pp_open_tag,
    Format.pp_close_tag, Format.open_tag, Format.close_tag,
    Format.formatter_tag_functions, Format.pp_set_formatter_tag_functions,
    Format.pp_get_formatter_tag_functions, Format.set_formatter_tag_functions,
    Format.get_formatter_tag_functions, Gc (mutability of the fields of type
    Gc.control), Lazy.lazy_from_fun, Lazy.lazy_from_val, Lazy.lazy_is_val,
    Obj.set_tag, Obj.truncate, Obj.final_tag, Obj.extension_constructor,
    Obj.extension_name, Obj.extension_id, Scanf.stdib, Scanf.fscanf,
    Scanf.kfscanf, Stdlib.( & ), Stdlib.( or ), String.set, String.copy,
    String.fill, String.unsafe_set, String.unsafe_fill, String.uppercase,
    String.lowercase, String.capitalize, String.uncapitalize, Thread.kill,
    Thread.wait_write, Thread.wait_read, the whole ThreadUnix module, the
    infix operator (.<-).
    (NicolĂĄs Ojeda BĂ€r, review by Damien Doligez)

  • (breaking change) #10896: Remove Stream, Genlex and Pervasives. Also remove legacy standalone
    bigarray library (the Bigarray module is now part of the standard library).
    (NicolĂĄs Ojeda BĂ€r, review by Kate Deplaix and Anil Madhavapeddy)

  • #10921: Use getentropy, when available, to seed the generator during
    (Michael Hendricks, review by Gabriel Scherer, Xavier Leroy, and
    Anil Madhavapeddy)
  • (breaking change) #10924: Add deprecated attribute to Printexc.catch, Printf.kprintf and
    (NicolĂĄs Ojeda BĂ€r, review by Damien Doligez)
  • #10986: Add Scanf.sscanf_opt, Scanf.bscanf_opt and Scanf.scanf_opt.
    (NicolĂĄs Ojeda BĂ€r, review by Florian Angeletti and Gabriel Scherer)
  • (breaking change) #11157: Rename “hash” in the “Hashtbl.SeededHashedType” signature to
    “seeded_hash”. This allows defining both seeded and unseeded hash functions in
    the same module.
    (NicolĂĄs Ojeda BĂ€r, review by Gabriel Scherer and Xavier Leroy)
  • #8878: Add String.hash and String.seeded_hash.
    (Tom Kelly, review by Alain Frisch and NicolĂĄs Ojeda BĂ€r)

Other libraries:

  • (breaking change) #9071, #9100, #10935: Reimplement Thread.exit() as raising the
    exception Thread.Exit, and mark Thread.exit as deprecated.
    The new implementation changes the behavior compared with
    the OCaml 4 implementation. For example, the new implementation
    causes Fun.finally finalizers to be run and catch-all exception
    handlers to prevent termination.
    (Jacques-Henri Jourdan and Xavier Leroy, review by Damien Doligez,
    Guillaume Munch-Maccagnoni, Gabriel Scherer, and Enguerrand Decorne)
  • #11034: Dynlink library, add a global lock to make dynlinking
    (Florian Angeletti, review by Gabriel Scherer)

  • #11087: deprecate Thread.wait_timed_read, Thread.wait_timed_write,
    Thread.select, Thread.wait_pid. The same functionality is available in the
    Unix module.
    (NicolĂĄs Ojeda BĂ€r, review by Anil Madhavapeddy)

  • #11309, #11424, #11427, #11545: Add Domain.recommended_domain_count.
    (Christiano Haesbaert, Konstantin Belousov, review by David Allsopp,
    KC Sivaramakrishnan, Gabriel Scherer, Nicolas Ojeda Bar)

  • #11461, #11466: Fix gethostbyaddr for IPv6 arguments and make it domain-safe
    (Olivier Nicole, NicolĂĄs Ojeda BĂ€r, David Allsopp and Xavier Leroy,
    review by the same)

  • #11479: Make Unix.symlink domain-safe on Windows
    (Olivier Nicole, review by Xavier Leroy and David Allsopp)


  • (breaking change) #11004: Litmus tests for checking the implementation of the
    memory model.
    (Luc Maranget, review by Gabriel Scherer and Xavier Leroy)
  • #11024: Handle alerts in ocamldoc.
    The alert [@@alert deprecated] is handled specifically and it’s no longer
    needed to duplicate the deprecated annotation (the alert for the compiler and
    the tag for the documentation). Every other alerts also appear in the
    (Jules Aguillon, review by Florian Angeletti)

  • #11065: Port the bytecode debugger to 5.0, adding support for effect handlers.
    (Damien Doligez and @fabbing, review by @fabbing and Xavier Leroy)

  • #11079: Add the -nobanners option to dumpobj.
    (SĂ©bastien Hinderer, review by Gabriel Scherer and Vincent Laviron)

  • #11100: Fix ocamltest to make sure failed tests are not counted as
    “unexpected error”.
    (Damien Doligez, review by SĂ©bastien Hinderer)

  • #11245: Merge the common code of ocamlcp and ocamloptp into a single module.
    (David Allsopp, review by SĂ©bastien Hinderer)

  • #11382: OCamlmktop use a new initialization module “OCamlmktop_init” to
    preserve backward-compatibility with user-module provided modules that install
    toplevel printers.
    (Florian Angeletti, review by Gabriel Scherer and David Allsopp)

Manual and documentation:

  • #11058: runtime/HACKING.adoc tips on debugging the runtime
    (Gabriel Scherer, review by Enguerrand Decorne and NicolĂĄs Ojeda BĂ€r)

  • #11093: Add an effect handlers tutorial
    (KC Sivaramakrishnan, review by François Pottier, Gabriel Scherer, François
    Bobot and Wiktor Kuchta)

  • #11192: Better documentation for condition variables.
    (François Pottier, review by Luc Maranget, Xavier Leroy, and Wiktor Kuchta)

  • #11093: Add tutorials on parallelism features and the relaxed memory model
    (KC Sivaramakrishnan, review by Damien Doligez, Anil Madhavapeddy, Gabriel
    Scherer, Thomas Leonard, Tom Ridge, Xavier Leroy, Luc Maranget, Fabrice
    Buoro, Olivier Nicole, Guillaume Munch-Maccagnoni, Jacques-Henri Jourdan)

  • #11640: Add Flambda commonly-used options to the ocamlopt manpage
    (Amandine Nangah, review by David Allsopp, Florian Angeletti,
    SĂ©bastien Hinderer, and Vincent Laviron)

  • #11676: Fix missing since annotation in the Sys and Format modules
    (Github user Bukolab99, review by Florian Angeletti)

  • #11813: Make new multicore chapters easier to discover, and emphasize impact
    on C bindings.
    (Edwin Török, review by KC Sivaramakrishnan, and Florian Angeletti)

Compiler user-interface and warnings:

  • #9140, #11131: New command-line flag -nocwd to not include implicit
    the current directory to the load path.
    (Thomas Roglin, review by Gabriel Scherer and NicolĂĄs Ojeda BĂ€r)

  • #11089: Add ‘since ’ information to compiler warnings.
    (André Maroneze, review by Florian Angeletti and Gabriel Scherer)

  • #10909: Disable warning 59 (assignment to immutable blocks) unless flambda
    invariant checks are enabled.
    (Vincent Laviron, review by Gabriel Scherer)

  • #10981, #11276: Implement a -cmi-file option for ocamlc and ocamlopt.
    (SĂ©bastien Hinderer, review by Damien Doligez, Daniel BĂŒnzli and
    Florian Angeletti)

  • (breaking change) #11049: Stop padding 1-digit compiler minor version numbers.
    (So for instance OCaml 5.0 rather than 5.00)
    (SĂ©bastien Hinderer, review by David Allsopp, Florian Angeletti and
    Xavier Leroy)
  • #11184, #11670: Stop calling ranlib on created / installed libraries
    (SĂ©bastien Hinderer and Xavier Leroy, review by the same)

  • #11253: Deprecate ocaml script and ocamlnat script where script has no
    extension and is an implicit basename.
    (David Allsopp, review by Florian Angeletti and SĂ©bastien Hinderer)

Internal/compiler-libs changes:

  • #10878, #10909: restore flambda after the Multicore merge.
    (Vincent Laviron, review by Gabriel Scherer and Xavier Leroy)

  • #10864, #10888: restore afl-fuzz mode for sequential programs.
    (Jan Midtgaard, review by Xavier Leroy and Gabriel Scherer)

  • #11008, #11047: rework GC statistics in the Multicore runtime
    (Gabriel Scherer, review by Enguerrand Decorne)

  • #11058: basic debugging documentation in runtime/HACKING.adoc
    (Gabriel Scherer, review by Enguerrand Decorne and NicolĂĄs Ojeda BĂ€r)

  • #11199: Stop installing topdirs.cmi twice. The toplevel now reads topdirs.cmi
    from +compiler-libs, as the debugger does.
    (David Allsopp, review by SĂ©bastien Hinderer)

  • #11007, #11399: META files for the stdlib, compiler-libs and other libraries
    (unix, dynlink, str, runtime_events, threads, ocamldoc) are now installed
    along with the compiler.
    (David Allsopp, Florian Angeletti, NicolĂĄs Ojeda BĂ€r and SĂ©bastien Hinderer,
    review by Daniel BĂŒnzli, Kate Deplaix, Anil Madhavapeddy and Gabriel Scherer)

Build system:

  • (breaking change) #10893: Remove configuration options --disable-force-safe-string and
    DEFAULT_STRING=unsafe as well as --enable-force-safe-string and
    DEFAULT_STRING=safe which are now the default unconditionally.
    (Kate Deplaix, review by Gabriel Scherer and David Allsopp)
  • #11092: Build native-code compilers on OpenBSD/aarch64.
    (Christopher Zimmermann, review by Anil Madhavapeddy)

  • #11126: Build system: make it possible to choose which ocamldep
    (and flags) to use when computing dependencies for the compiler.
    Add a -no-slash option to ocamldep to let users override -slash.
    (SĂ©bastien Hinderer, review by David Allsopp)

  • #11147: Factorize the stdlib-related compilation flags. Make it
    possible to control them consistently through the STDLIBFLAGS
    build variable. Make sure ocamldoc and ocamllex get compiled and
    linked with debugging information (-g).
    (SĂ©bastien Hinderer, review by Gabriel Scherer)

  • #11149: Make the bootstrap process reproducible on systems with non-big-endian
    floating point. If the boot/ artefacts are up-to-date, this means that running
    make bootstrap on any platform should not change the images in boot/ and paves
    the way for automated testing that the bootstrap is repeatable.
    (David Allsopp, review by Damien Doligez and SĂ©bastien Hinderer)

  • #11160: otherlibs: merge win32unix into unix.
    (SĂ©bastien Hinderer, review by David Allsopp, NicolĂĄs Ojeda BĂ€r,
    Xavier Leroy, Vincent Laviron and Antonin DĂ©cimo)

  • (breaking change) #11198, #11298: Install the Dynlink, Str and Unix libraries to individual
    subdirectories of LIBDIR. The compiler, debugger and toplevel automatically
    add -I +lib if required, but display an alert.
    (David Allsopp, review by Florian Angeletti, NicolĂĄs Ojeda BĂ€r,
    Valentin Gatien-Baron and SĂ©bastien Hinderer)
  • #11200: Install ocamlprof’s Profiling runtime module to a +profiling,
    removing it from the default namespace.
    (David Allsopp, review by SĂ©bastien Hinderer)

  • #11294: Switch minimum required autoconf to 2.71.
    (David Allsopp, review by Xavier Leroy)

  • #11370, #11373: Don’t pass CFLAGS to flexlink during configure.
    (David Allsopp, report by William Hu, review by Xavier Leroy and
    SĂ©bastien Hinderer)

  • #11487: Thwart FMA test optimization during configure
    (William Hu, review by David Allsopp and SĂ©bastien Hinderer)

  • #11097: Build native-code compilers on NetBSD/aarch64
    (Kate Deplaix, review by Anil Madhavapeddy)

Bug fixes:

  • #10768, #11340: Fix typechecking regression when combining first class
    modules and GADTs.
    (Jacques Garrigue, report by François Thiré, review by Matthew Ryan)

  • #10790: don’t drop variance and injectivity annotations when pretty printing
    with constraints (for example, with type +!'a t = ...).
    (Florian Angeletti, report by Luke Maurer, review by Matthew Ryan and
    Gabriel Scherer)

  • #11167: Fix memory leak from signal stack.
    (Antoni Ć»ewierĆŒejew, review by Gabriel Scherer and Enguerrand Decorne)

  • #11112: harden -use-runtime against spaces or quotes in the provided path
    (Gabriel Scherer, report by Brahima Dibassi, review by David Allsopp)

  • #11068, #11070: Fix typo in function name given in Unix_error exception for
    Unix.readlink on Windows.
    (David Allsopp, report by Xia Li-yao)

  • #10807: Don’t duplicate standard handles in the child process
    spawned by win32unix Unix.create_process if the handles were already
    inheritable. Fix broken signalling of EOF on standard handles if
    they were already inheritable.
    (Antonin DĂ©cimo, review by Xavier Leroy and NicolĂĄs Ojeda BĂ€r)

  • #10868: Fix off-by-1 bug when initializing frame hashtables
    (Jonah Beckford, review by Tom Kelly, NicolĂĄs Ojeda BĂ€r and
    KC Sivaramakrishnan)

  • #11077: Make dumpobj compatible with absence of naked pointer support
    (Olivier Nicole and Jan Midtgaard, review by Gabriel Scherer)

  • #11111: fix fork() usage in ocamltest C code.
    When calling fork() from C code with the Multicore runtime active,
    one needs to call caml_atfork_hook() on the forked child before it
    can use the OCaml runtime.
    (Gabriel Scherer, review by Xavier Leroy, report by Brahima Dibassi)

  • #10809: Use the WSA_FLAG_NO_HANDLE_INHERIT on Windows when creating
    sockets with WSASocket if the cloexec (non-inheritable) parameter is
    true. Fixes a race condition where a child process could inherit the
    socket and deadlock the parent.
    (Antonin DĂ©cimo, review by Xavier Leroy)

  • #11194, #11609: Fix inconsistent type variable names in “unbound type var”
    (Ulysse GĂ©rard and Florian Angeletti, review Florian Angeletti and
    Gabriel Scherer)

  • #11204: Fix regression introduced in 4.14.0 that would trigger Warning 17 when
    calling virtual methods introduced by constraining the self type from within
    the class definition.
    (NicolĂĄs Ojeda BĂ€r, review by Leo White)

  • #11263, #11267: caml/misc.h: check whether _MSC_VER is defined before using
    it to ensure that the headers can always be used in code which turns on
    -Wundef (or equivalent).
    (David Allsopp and NicolĂĄs Ojeda BĂ€r, review by NicolĂĄs Ojeda BĂ€r and
    SĂ©bastien Hinderer)

  • #11289, #11405: fix some leaks on systhread termination
    (Fabrice Buoro, Enguerrand Decorne, Gabriel Scherer,
    review by Xavier Leroy and Florian Angeletti, report by Romain Beauxis)

  • #11314, #11416: fix non-informative error message for module inclusion
    (Florian Angeletti, report by Thierry Martinez, review by Gabriel Scherer)

  • #11358, #11379: Refactor the initialization of bytecode threading,
    This avoids a “dangling pointer” warning of GCC 12.1.
    (Xavier Leroy, report by Armaël Guéneau, review by Gabriel Scherer)

  • #11387, module type with constraints no longer crash the compiler in presence
    of both shadowing warnings and the -bin-annot compiler flag.
    (Florian Angeletti, report by Christophe Raffalli, review by Gabriel Scherer)

  • #11392, #11392: assertion failure with -rectypes and external definitions
    (Gabriel Scherer, review by Florian Angeletti, report by Dmitrii Kosarev)

  • #11417: Fix regression allowing virtual methods in non-virtual classes.
    (Leo White, review by Florian Angeletti)

  • #11468: Fix regression from #10186 (OCaml 4.13) detecting IPv6 on Windows for
    mingw-w64 i686 port.
    (David Allsopp, review by Xavier Leroy and SĂ©bastien Hinderer)

  • #11482, #11542: Fix random crash in large closure allocation
    (Damien Doligez, report by Thierry Martinez and Vincent Laviron, review by
    Xavier Leroy)

  • #11508, #11509: make Bytes.escaped domain-safe
    (Christiano Haesbaert and Gabriel Scherer,
    review by Xavier Leroy,
    report by Jan Midtgaard and Tom Kelly)

  • #11516, #11524: Fix the deprecated_mutable attribute.
    (Chris Casinghino, review by NicolĂĄs Ojeda BĂ€r and Florian Angeletti)

  • #11576: Fix bug in Bigarray.Genarray.init in the the case of zero-dimensional
    (NicolĂĄs Ojeda BĂ€r, Jeremy Yallop, report by Masayuki Takeda, review by Jeremy
    Yallop and Florian Angeletti)

  • #11587: Prevent integer comparison from being used on pointers
    (Vincent Laviron, review by Gabriel Scherer)

  • #11622: Prevent stack overflow when printing a constructor or record
    mismatch error involving recursive types.
    (Florian Angeletti, review by Gabriel Scherer)

  • #11662, #11673: fix a memory leak when using Dynlink,
    the bug was only present in development version of OCaml 5.
    (Stephen Dolan, report by Andre Maroneze, review by Gabriel Scherer)

  • #11732: Ensure that types from packed modules are always generalised
    (Stephen Dolan and Leo White, review by Jacques Garrigue)

  • #11737: Fix segfault condition in Unix.stat under Windows in the presence of
    multiple threads.
    (Marc Lasson, NicolĂĄs Ojeda BĂ€r, review by Gabriel Scherer and David Allsopp)

  • #11776: Extend environment with functor parameters in strengthen_lazy.
    (Chris Casinghino and Luke Maurer, review by Gabriel Scherer)

  • #11533, #11534: follow synonyms again in #show_module_type
    (this had stopped working in 4.14.0)
    (Gabriel Scherer, review by Jacques Garrigue, report by Yaron Minsky)


Unbelievable. Thanks for the hard work y’all can’t wait to see where this goes and what we build with it!


Magnificent achievement!


Congratulations! I have some questions about plans to improve the debugging experience.

  • For the bytecode debugger, any plans to make it support multiple domains? From the source code I see that currently the debugger will abort when a new thread is created.

  • When debugging native code, any plans to make it so when debugging with gdb, (or other debugger) that the original ml source code is shown, and local variables can be displayed?


P.S. I’m experimenting with using the debugger when compiled to native. I see that sometimes the original OCaml code is shown, though part of the time it is showing the C runtime code. Perhaps some documentation on using a debugger with native code should be included in the manual.


Congratulations! Many people have been waiting for this for a long time.
Thanks to everyone who worked hard on this.

1 Like

Congratulations! It may have been a long time coming, but it’s paid off. This is a really nice design and implementation. I think it will lead to greater OCaml adoption, which benefits all of us, whether we use the new features or not.

1 Like

The source code is also available as a zipped file : https://codeload.github.com/ocaml/ocaml/zip/5.0.0

Where is hosted the ocamldoc for the stdlib of ocaml-5.0.0?
It seems that what I access online is always from some 4.* release.